需要从一个热编码形状中检索原始数据形状

Question

我收到一个数据集，其中包括先前转换为一个热编码的列。我想检索它的旧形状来做一些预处理和填充NAs方法，当然还有读取数据集的统计模型。

我得到的数据列:

<表类> team2_offensive_derived_var_0 team2_offensive_derived_var_1 team2_offensive_derived_var_2 team2_offensive_derived_var_3 team2_offensive_derived_var_4 team2_offensive_derived_var_5 team2_offensive_derived_var_6 team2_offensive_derived_var_7 team2_offensive_derived_var_8 team2_offensive_derived_var_9 team2_offensive_derived_var_10 tbody><<tr>0000000000100000000010000000001000000000100000000010000000001000000000100000000010000000001000000000100000000010000000000

您可以使用stack:

cols = ['row_id', 'team2_offensive_derived']
out = df.replace(0, pd.NA).stack().rename_axis(cols).reset_index()[cols]

输出:

row_id         team2_offensive_derived
0        0  team2_offensive_derived_var_10
1        1   team2_offensive_derived_var_9
2        2   team2_offensive_derived_var_8
3        3   team2_offensive_derived_var_7
4        4   team2_offensive_derived_var_6
5        5   team2_offensive_derived_var_5
6        6   team2_offensive_derived_var_4
7        7   team2_offensive_derived_var_3
8        8   team2_offensive_derived_var_2
9        9   team2_offensive_derived_var_1
10      10   team2_offensive_derived_var_0

有不同的列名:

out = (df
.replace(0, pd.NA)
.rename(columns=lambda x: x.replace('team2_offensive_derived_', ''))
.stack()
.rename_axis(cols)
.reset_index()[cols]
)

输出:

row_id team2_offensive_derived
0        0                  var_10
1        1                   var_9
2        2                   var_8
3        3                   var_7
4        4                   var_6
5        5                   var_5
6        6                   var_4
7        7                   var_3
8        8                   var_2
9        9                   var_1
10      10                   var_0

使用说明:

#split by last previous _
df.columns = df.columns.str.rsplit('_', 2, expand=True)
#replace 0 to NaNs, so reshape remove rows with 0
df = df.replace(0, np.nan).stack([1,2])
#join MultiIndex value
df.index = df.index.map(lambda x: f'{x[1]} {x[2]}')
#create DataFrame
df = df.index.to_frame(name='team2_offensive_derived',index=False)
print (df)
team2_offensive_derived
0                   var 10
1                    var 9
2                    var 8
3                    var 7
4                    var 6
5                    var 5
6                    var 4
7                    var 3
8                    var 2
9                    var 1
10                   var 0